CAF is the Apple container that has no size limit, unlike WAV. MP3 is MPEG-1 Audio Layer III, lossy, and the one audio format that every device made since 1995 can play. ffmpeg sits between the two, and this page says exactly what happens to a CAF file on the way to becoming an MP3 one.
What runs when a CAF file becomes MP3
ffmpeg held to a single thread, on a machine we rent and watch. Your CAF file is uploaded once, ffmpeg runs once, the MP3 comes back, and neither file is kept. CAF to MP3 is one of the 3649 pairs that engine was probed on with a real file, which is why it has a page here and why the pairs the probe could not prove do not.
What survives from the CAF into the MP3
The running time and the channel layout. A stereo CAF comes back as a stereo MP3 of the same length: this is a re-encode, not an edit, and nothing here normalises, trims or boosts the CAF.
The whole of the CAF, start to end. MP3 is a lossy encoding, but it encodes all of the CAF rather than a shortened or quieter version of it, and the loudness you hear is the loudness you sent.
What CAF to MP3 costs you
Exactness, once and for good. The CAF holds every sample; MP3 keeps what its model believes you can hear and discards the rest. Going back from MP3 to CAF later rebuilds the file size and none of the sound, which is the single most common misunderstanding about this pair.
Album art, where the CAF carried any. ffmpeg passes -vn on every audio target here, which counts an embedded cover as a video stream and drops it along with the film it was never part of. That is deliberate rather than careless: keeping it makes the mux fail outright on several audio containers, and hiding the choice would be lying by omission.
Any hope of MP3 sounding better than the CAF. Nothing in this pass adds information, and a bitrate higher than the CAF deserved only makes a larger file of the same recording.
Where CAF and MP3 files come from
CAF. Core Audio Format, Apple's container. Its reason to exist is that WAV cannot exceed four gigabytes and a long recording can: CAF carries the same samples with a header that counts higher.
If a model is going to read the MP3
An MP3 file is a waveform, and a language model reads text, so handing an assistant this MP3 directly gets you nothing. The MP3 becomes useful the moment somebody turns the sound into words, and that is a separate job with its own page here. CAF to MP3 is the step before it: get the sound out of the CAF in a container the transcriber accepts, then hand the model a transcript instead of a file it cannot open.
Running CAF to MP3 against the real engine with a real file, ffmpeg wrote 24,468 bytes of MP3 in 0.1 seconds, machine otherwise idle. That is one CAF file on one day and not an average, which is why the number is given together with the file that produced it.
The CAF to MP3 verdict was reached by ffprobe, meaning the container was actually demultiplexed, recognised under the right name, and carried at least one stream. Which check was used matters, because they do not all prove the same thing, and a status code of 200 proves nothing whatsoever about whether the MP3 file has anything inside it.
The same CAF file, sent somewhere else
Other outputs the probe measured out of a CAF file, so the cost of choosing MP3 can be read against something. Sizes do not compare across engines, because each family was probed with its own CAF witness file.
CAF to WAV: 264,678 bytes in 0.1 seconds, verified by ffprobe.
CAF to FLAC: 44,337 bytes in 0.1 seconds, verified by ffprobe.
CAF to OGG: 8,670 bytes in 0.1 seconds, verified by ffprobe.
CAF to M4A: 27,633 bytes in 0.2 seconds, verified by ffprobe.
CAF to OPUS: 27,591 bytes in 0.1 seconds, verified by ffprobe.
CAF to AAC: 27,219 bytes in 0.2 seconds, verified by ffprobe.
Other ways into MP3, and what they measured
Among the published routes into MP3, CAF is the number 17 largest output of the 22 measured. The witness files differ, so this ranks the probe run and not your document.
3GP to MP3: 24,780 bytes in 0.1 seconds.
MPG to MP3: 24,468 bytes in 0.1 seconds.
OPUS to MP3: 24,429 bytes in 0.2 seconds.
WAV to MP3: 24,468 bytes in 0.2 seconds.
M4A to MP3: 24,776 bytes in 0.1 seconds.
What we will not pretend about CAF to MP3
A CAF file over 150 MB is refused before the upload finishes rather than after it, so you do not wait for a rejection.
A CAF to MP3 run that passes 240 seconds is killed, and the ffmpeg process is killed with it. A run left behind would sit on one of the machine's two cores until somebody noticed.
The CAF file behind the measurement on this page was a 3 second sine wave at 440 hertz, built by ffmpeg for the probe itself. That is why CAF to MP3 came out at 0.1 seconds, and it is not what your file will do. The measurement proves the pair works and says almost nothing about how long yours will take.
CAF to MP3 costs half a credit per conversion, whatever the length of the CAF. That is the same as an image or a document here, and it is a measurement rather than a gesture: an audio target on this engine is done in under two seconds, where a video target can take minutes.
A CAF longer than 60 minutes is refused before encoding. ffprobe measures the CAF first and answers with a refusal, rather than starting a job the 240 second guard would kill halfway to the MP3.
CAF to MP3 waits at the audio door, which is 2 slots wide and gives up after 5 seconds rather than making you wait twenty times the length of the work. Until the 30th there was one door for everything, and a single video transcode starved every audio conversion for its whole duration: an MP3 that needed less than a second was answered with a refusal after twenty.
Nothing here reads what is inside the CAF. No transcription, no scene detection, no summary: CAF to MP3 is a container and codec operation, and the MP3 says exactly what the CAF said, in a different wrapper. The sound is the whole payload.
One CAF file at a time, chosen in the browser. There is nothing else to set up and nothing else on offer.
CAF to MP3: what people ask
What actually converts my CAF file to MP3?
ffmpeg does it, held to a single thread, on a machine we rent and watch. Not a browser trick and not somebody else service: the CAF file is uploaded once, ffmpeg runs once, the MP3 comes back, and neither file is kept afterwards.
How long does CAF to MP3 take?
The probe measured 0.1 seconds for 24,468 bytes of MP3, on a 3 second file. An audio target scales far better than a video one: three minutes of source through this engine takes under a second, because only the sound is encoded. Past 240 seconds any run is stopped, and an audio one does not get close.
What do I lose going from CAF to MP3?
The one to know about first: Exactness, once and for good. The CAF holds every sample; MP3 keeps what its model believes you can hear and discards the rest. Going back from MP3 to CAF later rebuilds the file size and none of the sound, which is the single most common misunderstanding about this pair.
Will the MP3 sound better than the CAF?
No. Nothing in this pass adds information to the CAF, so the MP3 is at best an exact carry of what you sent and at worst one more lossy pass over it. Asking for a higher bitrate than the CAF deserved just makes a bigger file of the same recording.
Is CAF to MP3 free?
There is a free allowance every month, and one CAF to MP3 conversion costs half a credit against it however long the CAF runs. An audio target is charged per conversion and not per minute, because the engine encodes three minutes of sound in well under a second.